MSc in Mathematics and Finance, Imperial College London

Dr Robert Nrnberg u

Computing in C++ Part II : Object Oriented Programming in C++

Dr Robert Nrnberg u

What is OOP?
Object oriented programming is a fairly new way to approach the task of programming. It supersedes the so called procedural or structured programming languages like Algol, Pascal or C, that have been around since the 1960s. The essence of structured programming is to reduce a program into smaller parts and then code these elements more or less independently from each other. Although structured programming has yielded excellent results when applied to moderately complex programs, it fails when a program reaches a certain size. To allow for more complex programs to be written, the new approach of OOP was invented. OOP, while allowing you to use the best ideas from structured programming, encourages you to decompose a problem into related subgroups, where each subgroup becomes a self-contained object that contains its own instructions and data that relate to that object. In this way complexity is reduced, reusability is increased and the programmer can manage larger programs more easily. All OOP languages, including C++, share the following three capabilities: Encapsulation the ability to dene a new type and a set of operations on that type, without revealing the representation of the type. Inheritance the ability to create new types that inherit properties from existing types. Polymorphism the ability to use one name for two or more related but technically dierent purposes; one interface, multiple methods.

Recommended books and sites

Daniel Duy, Introduction to C++ for Financial Engineers : An Object-oriented Approach, 2006 Steve Oualline, Practical C++ Programming, 1995 Herbert Schildt, Teach yourself C++, 1992 Jesse Liberty, Teach yourself C++ in 24 hours, 1999 Bruce Eckel, Thinking in C++, Online at www.mindview.net/Books/TICPP/ThinkingInCPP2e.html Bjarne Stroustrup, The C++ Programming Language, 1997 (technical) www.research.att.com/ bs/C++.html (Stroustrup: C++) www.cplusplus.com www.cppreference.com

Compilers
Windows: Microsoft Visual C++ .NET with Integrated Development Environment (IDE) GNU C++ compiler (g++) as part of Cygwin or MinGW or, without IDE Dev-C++ free compiler/IDE that is GNU compatible

Linux:

GNU C++ compiler (g++) part of any distribution Intel C++ compiler (icc) free for students

Mac: Xcode free compiler/IDE that is GNU compatible Windows/Linux/Mac: Code::Blocks free compiler/IDE that is GNU compatible NetBeans free compiler/IDE that is GNU compatible

v/2012/02/20

OOP in C++

Dr Robert Nrnberg u

Contents
1 Revision 1.1 Hello World! revisited . . . 1.2 Using C header les . . . . . 1.3 Macros . . . . . . . . . . . . . 1.4 Enumerations . . . . . . . . . 1.5 Strings . . . . . . . . . . . . . 1.6 Signatures of functions . . . . 1.7 Passing variables to functions 1.7.1 Reference vs. pointers 1.7.2 Reference vs. value . . 1.7.3 Keyword const . . . . 1.7.4 Default arguments . . 1.8 File input and output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 4 4 4 5 5 5 6 6 6 7 7 7 8 8 8 9 9 10 10 10 11 11 11 12 12 12 13 14 14 14 14 15 16 16 17 17 18 19 19 19 20 21 22 22 23 23 24 25

2 Objects 2.1 Structures . . . . . . . . . . . . . . . . 2.1.1 Pointers to structures . . . . . 2.2 Classes private and public members 2.3 Methods . . . . . . . . . . . . . . . . . 2.4 Constructors . . . . . . . . . . . . . . 2.4.1 Copy constructors . . . . . . . 2.5 Destructors . . . . . . . . . . . . . . . 2.6 The this pointer . . . . . . . . . . . . 2.7 Keyword const . . . . . . . . . . . . . 2.8 Classes inside classes . . . . . . . . . . 2.9 Friend functions . . . . . . . . . . . . 2.10 Friend classes . . . . . . . . . . . . . . 2.11 Header les . . . . . . . . . . . . . . . 2.12 Strings revisited . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

3 Overloading 3.1 Pointers to overloaded functions . . . . . 3.2 Overloading constructors . . . . . . . . . 3.3 Operator overloading . . . . . . . . . . . 3.4 Overloading binary operators . . . . . . 3.4.1 Overloading relational operators 3.5 Overloading unary operators . . . . . . 3.6 Overloading the () operator . . . . . . . 3.7 Using friend operators . . . . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

4 Inheritance 4.1 Protected members . . . . . . . . . . . . . 4.2 Constructors, destructors and inheritance 4.3 Pointers to derived classes . . . . . . . . . 4.4 Virtual functions . . . . . . . . . . . . . . 4.4.1 Pure virtual functions . . . . . . . 4.4.2 Abstract classes . . . . . . . . . . . 4.4.3 Constructors and destructors . . . 4.4.4 Pure virtual destructors . . . . . . 4.4.5 Inheriting virtual functions . . . . 4.5 Friend functions and inheritance . . . . . 4.6 Example for inheritance . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

OOP in C++

Dr Robert Nrnberg u

5 Templates 5.1 Function templates . . . . . . . 5.2 Class templates . . . . . . . . . 5.2.1 Inheritance . . . . . . . 5.2.2 Template specialization 5.2.3 Partial specialization . . 5.3 Compile time computations . . 6 STL 6.1 Vectors . . . . . . . . . . . 6.2 Other sequential containers 6.3 Associative containers . . . 6.4 Iterators . . . . . . . . . . . 6.5 Functors . . . . . . . . . . . 6.6 Algorithms . . . . . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

25 25 26 26 27 28 29 29 30 30 31 32 32 33 34 34 35 35 36 36 37 37

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

7 Miscellaneous 7.1 Namespaces . . . . . . . . . . . 7.2 Advanced I/O . . . . . . . . . . 7.3 Static class members . . . . . . 7.4 Conversion functions . . . . . . 7.5 Exception handling . . . . . . . 7.6 Temporary objects . . . . . . . 7.7 Passing functions as arguments

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

OOP in C++

Dr Robert Nrnberg u

Revision

Before we embark on the topic of object oriented programming, lets remind ourselves of some other important features of C++.

1.1

Hello World! revisited

#include <iostream> using namespace std; int main() { cout << "Hello World!" << endl; return 0; }

In C++ this standard example should look like

The rst two lines of the above example show the new ANSI-C++ compliant way to include the standard C++ libraries. Although many compilers still support the old style, eventually all C++ compilers will only accept the ANSI standard. Note that without the second line the code would have to look something like
#include <iostream> using namespace std; int main() { std::cout << "Hello World!" << std::endl; return 0; }

See 7.1 below for further details on namespaces.

1.2

Using C header les

All of the old C standard header les, like stdlib.h, math.h, etc. can still be used. This should be done as follows.
#include <iostream> #include <cmath> using namespace std; // was: math.h in C

int main() { cout << "exp(1.0) = " << exp(1.0) << endl; return 0; }

For more details on the ANSI-C++ way to include header les from the standard library visit www.cplusplus.com/doc/ansi/hfiles.html.

1.3

Macros

The C/C++ compiler evaluates certain preprocessing directives before actually starting to compile a source code. These preprocessing directives are lines in your program that start with #. An example is the #include statement, which tells the compiler what kind of system calls and libraries you want to use in your program. A macro is another preprocessing directive. In its simplest form, a macro is just an abbreviation. It is a name which stands for a fragment of code. Usually, this will be some specic constant, such as Pi and DEBUG in the example below. Before you can use a macro, you must dene it explicitly with the #define directive. It is followed by the name of macro and then the code it should be an abbreviation for. Macros can also be dened to take arguments. In this case, any expansion you dene for the macro, will be applied with the arguments substituted by the values you are giving. The function macros SQR() and MAX() in the example below demonstrate the usage. Finally, you can also use conditional directives that allow a part of the program to be ignored during compilation, on some conditions. A conditional can test either an arithmetic expression or whether a 4

OOP in C++

Dr Robert Nrnberg u

name is dened as a macro. Check the usage of #if, #else, #endif, and #ifdef, #ifndef in the example below.
#include <iostream> using namespace std; #define SQR(a) ((a)*(a)) #define MAX(a,b) ((a) > (b) ? (a) : (b)) #ifndef Pi #define Pi 3.1415926535897932384626433832795028841971 #endif #define DEBUG 1 int main() { #ifdef MAX if (MAX(2.0,3.4) > 2.5) cout << "Square of pi is: " << SQR(Pi) << endl; #endif #if DEBUG cout << "Debugging switched on." << endl; cout << " --- " << __FILE__ << " line " << __LINE__ << "--<< __DATE__ << "." << endl; #else cout << "Debugging switched off." << endl; #endif return 0; }

Compiled on "

Note that the example also uses some standard predened macros. They are available with the same meanings regardless of the machine or operating system on which you are using C/C++. Their names all start and end with double underscores, e.g. __FILE__, __LINE__ and __DATE__. More information on the C/C++ preprocessor can be found at doc.ddart.net/c/gcc/cpp toc.html.

1.4

Enumerations

As in C, enumerations provide a convenient way to dene variables that have only a small set of meaningful values. Here is a simple example.
enum colour {red, green, blue, black}; colour foreground = black;

1.5

Strings

C++ may use the C string functions, where a string is represented by a char*, but they rely on a null termination and proper memory allocation to hold the string. The C++ string class attempts to simplify string manipulation by automating much of the memory allocation and management.
#include <iostream> #include <string> using namespace std; int main() { string a("abcd efg"); string b = "xyz ijk"; string c = a + b; // concatenation cout << "First character of "<< c <<" is : " << c[0] << endl; // as in C return 0; }

1.6

Signatures of functions

In C++, in contrast to C, it is possible that two dierent functions share exactly the same name. That is because in C++ a function is identied by its signature which comprises both the functions name and 5

OOP in C++

Dr Robert Nrnberg u

the list of arguments it takes. Hence the following code is perfectly ne.
#include <iostream> using namespace std; void add_some(int *i) { (*i) += 3; } void add_some(double *x) { (*x) += 1.5; } int main() { int a(1); double b(1.23); add_some(&a); add_some(&b); cout << "a = " << a << ", b = " << b << endl; return 0; }

// a = 4, b = 2.73

This feature of C++ is also referred to as function overloading. More on that in 3.

1.7
1.7.1

Passing variables to functions

Reference vs. pointers

Passing a variable to a function by reference or by pointer is essentially the same thing. For instance, the following two functions do exactly the same.
#include <iostream> using namespace std; void add_one_pt(double *x) { ++(*x); } void add_one_rf(double &x) { ++x; } int main() { double a(1.23), b(0.0); add_one_pt(&a); add_one_rf(b); cout << "a = " << a << ", b = " << b << endl; return 0; }

However, there are times when we have to use pointers in order to pass a variable to a function. In particular, if we cannot be sure that the referred to variable will always be dened, we have to pass it via a pointer. Here is a simple example of such an instance.
#include <iostream> using namespace std; void execute(void (*f)()) { if (f != NULL) f(); else cout << "Nothing to execute!" << endl; } void print() { cout << "Hello World!" << endl; } int main() { void (*p)() = print; execute(NULL); execute(p); return 0; }

1.7.2

Reference vs. value

When we do not want to change the value of a variable that is given as an argument to a function, we usually pass it to the function by value. This causes the function to create a local copy of the passed 6

OOP in C++

Dr Robert Nrnberg u

variable. When passing large data structures this is usually undesirable. A way out is to pass the variable by reference using the keyword const. This ensures that the compiler will not allow any change to the passed variable within the function. The following two functions demonstrate the two concepts.
#include <iostream> using namespace std; void print_by_val(double x) { cout << x << endl; } void print_by_ref(const double &x) { cout << x << endl; } int main() { double p = 3.14; print_by_val(p); print_by_ref(p); return 0; }

1.7.3

Keyword const

Similar to the above usage of const in connection with passing arguments by reference, it can be employed when passing arguments by pointers as well. 1.7.4 Default arguments

C++ oers the possibility to implement default arguments for functions, in case that they are called with less parameters than originally specied. Here is a simple example.
#include <iostream> using namespace std; void print(int i = 10, int j = 5) { cout << i << " " << j << endl; } int main () { print(2); return 0; }

// 2 5

Note that all default parameters must be to the right of any parameters that do not have defaults. Also, once you dene a default parameter, all the following parameters have to have a default as well.

1.8

File input and output

Input and output from and to a le needs the header le fstream and is similar to using cin and cout. E.g.
#include <fstream> using namespace std; int main() { ofstream fout("output.txt"); fout << "Hello World!" << endl; fout.close(); return 0; }

When dealing with variable length input les, it should be noted that the insertion operator >> itself returns true whenever the last input operation has been successful. See the example below for an illustration.
#include <iostream> #include <fstream> using namespace std; int main() { ifstream fin("data.dat"); if (! fin.is_open()) {

OOP in C++

Dr Robert Nrnberg u

cout << "Error opening the file data.dat." << endl; return -1; } double mean = 0.0, x; int n = 0; while ( fin >> x ) { mean += x; n++; } fin.close(); if (n > 0) mean /= n; cout << "The mean of the " << n << " numbers is " << mean << "." << endl; return 0; }

Objects

The term object in C++ refers to an instance of a class. In the example below, studlist[0] is an instance of the struct student, where we have noted that a struct (as known from C) is simply a class with special properties, see 2.2 for details. The support for classes is the most signicant dierence between C and C++. A class is essentially an abstract data type. It is a collection of data together with functions that operate on the data. We will refer to these as member data and member functions of a class, respectively. By declaring a class we can create a new data type that, if we are careful, is as powerful as any of the basic types. We start our considerations with a simpler concept known from C.

2.1

Structures
#include <iostream> #include <string> using namespace std; struct student { string name, firstname; int year; }; int main() { int N; cout << "Number of students: "; cin >> N; student *studlist = new student[N]; for (int i = 0; i < N; i++) { cin >> studlist[i].name >> studlist[i].firstname >> studlist[i].year; } // etc return 0; }

Structures allow us to bundle data that naturally belong together into one object. E.g.

Note that a struct denes a type that can be used (almost) as freely as any of the basic types, such as double, int, bool etc. In particular, we can pass variables of this newly dened type to functions, use references and pointers to and declare arrays with these variables. 2.1.1 Pointers to structures

At times it may be preferable to use (arrays of) pointers to structures. In this case the main() program above would have to be changed to something like

OOP in C++

Dr Robert Nrnberg u

int main() { int N; cout << "Number of students: "; cin >> N; student **studlist = new student* [N]; for (int i = 0; i < N; i++) { studlist[i] = new student; cin >> studlist[i]->name >> studlist[i]->firstname >> studlist[i]->year; } // etc return 0; }

Observe that here the members of a struct are accessed via e.g. pointer->firstname, whereas before we used variable.firstname.

2.2

Classes private and public members

As mentioned earlier, classes and structures are almost the same. In fact, the only dierence is that by default the members of a structure are accessible from the outside via the . and -> operators. In a class, on the other hand, the members are by default hidden within the class and cannot be accessed from the outside. We say that the members of a class are declared private by default. However, if we want to, we can declare members of a class to be public. Hence the struct student in 2.1 can equivalently be dened as
class student { public: string name, firstname; int year; };

Nevertheless, it is somewhat bad programming practice to declare member data to be public. For a safe and easy to update code one should always declare these as private. But if the private members of a class are not accessible outside the class, we have to nd a way to change and inspect their values. This can be done with the help of functions that have access to the member data and are themselves declared inside the class.

2.3

Methods

Member functions of a class are also referred to as methods. We have come across an instance of this already during le input and output, cf. 1.8, where the line fout.close(); called the member function close() for the instance fout of the class ofstream. If we make the member data of our student class private, we should provide functions to create and print the data, say. The declaration of the class would then look as follows.
class student { string name, firstname; int year; public: void create(const string &n, const string &fn, int y); void print(); };

Observe that the member data is declared private, since this is the default setting for a class. So far we have only declared the new member functions. Before we can actually use them, we have to give their denitions. This can be done as follows.
void student::create(const string &n, const string &fn, int y) { name = n; firstname = fn; year = y; } void student::print() { cout << name << ", " << firstname << " (YEAR " << year << ")" << endl; }

Note the usage of the scope operator :: in order to dene the member functions. Here is an example of how to use the newly dened methods. 9

OOP in C++

Dr Robert Nrnberg u

int main() { student s1; s1.create("Smith", "Alan", 1); s1.print(); return 0; }

The big advantage of dening all the member data as private is that rstly we can easily trace an error if something goes wrong, since the only place where the data is changed is in some of the member functions. And secondly, it is very easy to enrich and modify the class later on. For example, we could choose to include a variable address in the class. All we have to do in order to make sure that this new variable is always properly initialized in our program is to change the create method to something like
void student::create(const string &n, const string &fn, int y) { name = n; firstname = fn; year = y; address = "unknown"; }

and maybe provide another create method to initialize all four variables, e.g.
void student::create(const string &n, const string &fn, int y, const string &a) { name = n; firstname = fn; year = y; address = a; }

2.4

Constructors

In order to properly initialize an object, C++ allows to dene a constructor function that is called each time an object of that class is created. A constructor function has the same name as the class of which it is a part and has no return type. On recalling default arguments from 1.7.4, the constructor for the original student class can be dened as follows.
class student { string name, firstname; int year; public: student(const string &n = "unknown", const string &fn = "unkown", int y = 1) { name = n; firstname = fn; year = y; } void print(); };

2.4.1

Copy constructors

A constructor that takes as its only argument an object of its own class is called copy constructor. It is also used whenever a parameter is passed by value and when an object is returned by a function. If there is no copy constructor dened for the class, C++ uses the default built-in copy constructor, which creates a bit wise copy of the object. Hence one only needs to dene this special constructor if the class has a pointer to dynamically allocated memory. The declaration of the copy constructor in the case of the student class would have to look as follows.
class student { public: student(const student &s); // etc };

Observe that we have to pass the parameter by reference, since otherwise a copy of the argument would have to be made, which would invoke the copy constructor, for which another copy would have to be made . . . etc. Note also that we used the keyword const to ensure that the argument is not changed inside the function.

2.5

Destructors

The complement of a constructor is a destructor. This function is called automatically whenever an object is destroyed, i.e. for local variables when they go out of scope and for global variables when the program ends. This special member function should never be called explicitly. A destructor needs to be

10

OOP in C++

Dr Robert Nrnberg u

implemented only for classes that have a pointer to dynamically allocated memory, for all other classes the default destructor is sucient. The destructor function has the name of the class preceded with a ~. It has no return type and takes no parameters. Here is an example.
class array { double *a; int len; public: array(int l = 0) { len = l; a = new double[len]; } ~array() { delete[] a; } };

// constructor // destructor

2.6

The this pointer

C++ contains a special pointer that is called this. this is a pointer that is automatically passed to any member function when it is called, and it is a pointer to the object that generates the call. E.g. given the statement foo.bar(), the function bar() is automatically passed a pointer to foo, which is the object that generates the call. This pointer is referred to as this. Hence an equivalent denition of the function student::print() from 2.3 is
void student::print() { cout << this->name << ", " << this->firstname << " (YEAR " << this->year << ")" << endl; }

Of course, there is no reason why one should use the this pointer as shown here, because the shorthand form is much easier. However, the this pointer has several uses, including aiding in overloading operators. See e.g. 3.4 for more details.

2.7

Keyword const

When passing objects by using the keyword const, as described in 1.7.3, as a parameter to a function that possibly calls some of the objects member functions, one has to tell C++ that these methods will not change the passed object. To do this one has to insert the keyword const in the methods declarations. Here is a short example.
class myclass { int a; public: void showvalues(const myclass &A) { this->print(); A.print(); } void print() const { cout << a << endl; } // const needed! };

It is always advisable to use the keyword const for member functions that you as the programmer know will not change the objects data. For instance, it makes debugging easier and it helps to avoid some errors already at the compilation stage.

2.8

Classes inside classes

Apart from member data and member functions, classes can also dene their own data types. These new class denitions, inside the given class, can then be used by the classs member functions as well as from outside the class. In the latter case, one has to use the scope operator. See the following example for details.
#include <iostream> using namespace std; class myclass { int a; public: class local_type { double t; public: local_type(double x) { t = x; }

11

OOP in C++

Dr Robert Nrnberg u

void print() const { cout << "Local type: " << t << endl; } }; void print() const { cout << a << endl; } }; int main() { myclass::local_type z(2.3); z.print(); return 0; }

Note that the same access restriction for member data and member functions also applies to classes dened inside other classes. I.e. these classes can only be used from outside the class in which they are dened, if they were declared as public.

2.9

Friend functions

There will be times when the programmer wants a function to have access to the private members of a class without that function actually being a member of that class. To this end, C++ supports friend functions. A friend is not a member of a class but still has access to its private elements. Since a friend function is neither private nor public, it can be declared anywhere in the class denition. The following example demonstrates the syntax as well as how to make a function friend of more than one class.
class classB; class classA { int a; friend bool is_equal(const classA &, const classB &); }; class classB { int b; friend bool is_equal(const classA &, const classB &); }; bool is_equal(const classA &A, const classB &B) { return (A.a == B.b); }

As a general rule of thumb, use member functions where possible, and friend functions only where necessary. In much the same way as you should make data and methods of your classes private where possible, and public where necessary.

2.10

Friend classes

In much the same way that we can make a (global) function a friend of a class, we can also declare another class to be a friend of a class. This means that the member functions of the friend class are allowed to access also the private member data of the original class. Here is a short example.
class classA { int a; friend class classB; }; class classB { int b; public: void convert(const classA &A) { b = A.a; } };

2.11

Header les

In large programming projects it is often necessary to make use of header les. Usually, the header le will include all the information, that other parts of the project need to know in order to use certain functions and classes. Including a header le with the #include preprocessing directive, followed by the

12

OOP in C++

Dr Robert Nrnberg u

name of the header le in inverted commas, is like copying and pasting the contents of the header le into the C++ source le. Below you can nd a simple example. Contents of "script_header.h"
#ifndef SCRIPT_HEADER_H #define SCRIPT_HEADER_H class A { int a; public: A(int = 0); void print() const; private: void check() const; }; #endif // SCRIPT_HEADER_H

Contents of "script_header.cc"
#include <iostream> #include "script_header.h" using namespace std; A::A(int i) { a = i; } void A::print() const { cout << a << endl; check(); } void A::check() const { if (a==5) cout << "5!!" << endl; } int main() { A x(2), y(5), z; x.print(); y.print(); z.print(); return 0; }

Note that the macro denition inside the header le prevents the class to be dened twice, in case the header le is included in more than one source le. Note furthermore, that a header le usually only provides the declaration of member functions, whereas the actual implementation takes places in the C++ source le.

2.12

Strings revisited

Now that we are familiar with the concept of class member functions, we can have another look at the C++ string class. In 1.5 we already saw how to use the basic features of the C++ string class. In the following program we demonstrate the use of some other useful member functions of this class. For further details visit e.g. www.cppreference.com/cppstring.html.
#include <iostream> #include <string> using namespace std; int main() { string f(" Leading and trailing blanks "); cout << "String f : " << f << endl; cout << "String length : " << f.length() << endl; cout << "String f : " << f.append("ZZZ") << endl; cout << "String length : " << f.length() << endl; cout << "substr(5,9) : " << f.substr(5,9) << endl; cout << "assign(\"xyz\",0,3) : " << f.assign("xyz",0,3) << endl; cout << "insert(1,\"abc\") : " << f.insert(1,"abc ") << endl; string g("abc abc abd abc"); cout << "String g : " cout << "replace(12,1,\"xyz\") : cout << "replace(4,3,\"xyz\",2): cout << "replace(4,3,\"ijk\",1): cout << "find(\"abd\",1) : cout << "find(\"xyzb\") : return 0; }

<< g " << " << " << " << " <<

<< endl; g.replace(12,1,"xyz") << endl; g.replace(4,3,"xyz",2) << endl; g.replace(4,3,"ijk",1) << endl; g.find("abd",1) << endl; g.find("xyzb") << endl;

13

OOP in C++

Dr Robert Nrnberg u

The above code produces the following output:

String f : String length : String f : String length : substr(5,9) : assign("xyz",0,3) : insert(1,"abc") : String g : replace(12,1,"xyz") : replace(4,3,"xyz",2): replace(4,3,"ijk",1): find("abd",1) : find("xyzb") :

Leading and trailing blanks 37 Leading and trailing blanks 40 eading an xyz xabc yz abc abc abd abc abc abc abd xyzbc abc xy abd xyzbc abc iabd xyzbc 5 9 ZZZ

A further important member function is string::c_str() which returns a pointer to a character array with the same characters as the string that generates the call. This method is needed if you want to use some older library routines such as sscanf.

Overloading

As already indicated in 1.6 and 1.7.4, in C++ it is possible to dene several functions with the same name, provided that the type of their arguments or the number of the arguments they take dier. After classes, this is perhaps the next most important feature of C++.

3.1

Pointers to overloaded functions

Sometimes it might be necessary to pass the address of an overloaded function to a pointer. E.g. when passing a function as a parameter to another function. The way to tell C++ which functions address we would like to access is to include the list of parameters that the function takes into the denition of the pointer. Below is a simple example.
#include <iostream> using namespace std; void print(int i) { cout << "i = " << i << endl; } void print(double a, double b) { cout << "a = " << a << ", b = " << b << endl; } int main() { void (*p)(int) = print; void (*q)(double, double) = print; p(2); q(1.0,2.0); return 0; }

// //

i = 2 a = 1, b = 2

3.2

Overloading constructors

As we have seen in 2.4, a very common example of function overloading is to provide several dierent versions of a classs constructor. This is only natural, since a class needs a constructor for each way that an object of that class is declared in a program. In particular, when dynamically allocating memory for objects of a class, one has to dene a default constructor that takes no arguments. Here is an example for the student class.
student *p; p = new student[1]; p->print(); delete [] p; // calls student::student(); // unknown, unkown (YEAR 1)

3.3

Operator overloading

In C++, operators like +, *, <, >>, [] and () are just a very special type of functions. However, when overloading operators some special rules apply. In particular, it is not possible to change the number of arguments an operator takes. Furthermore, it is not possible to redene an operator for the basic 14

OOP in C++

Dr Robert Nrnberg u

types. I.e. when overloading an operator at least one of its arguments must be a newly dened class (or enumeration). Hence an overloaded operator never loses its original meaning. Instead, it gains additional meaning relative to the class for which it is dened. Furthermore, even if an operator is overloaded, it maintains its precedence over other operators and its associativity. When overloading an operator relative to a newly dened class, where an object of that class appears on the left hand side of the operator, there are two ways of dening the operator function. Either the operator is overloaded by a member function of the class, or the operator is dened as a global function. In the rst case the operator function has to be passed one parameter less than the operator takes, since the call of the operator function will always be generated by the object on the left hand side of the operator. In the second case, the operator function takes as many parameters as the operator itself. Moreover, in general this global operator function will be declared as a friend of the class it is dened for, cf. 3.7. Finally, operators may not have default parameters. In the following example, the [] subscript operator of a class is overloaded.
#include <iostream> using namespace std; class A { public: void operator[] (int i) { cout << "You want: [" << i << "]." << endl; } }; int main() { A a; a[5]; return 0; }

// overloading A[int]

// You want: [5].

Although it is permissible to have an operator function perform virtually any action (see example above), it is highly advisable to stick to operations similar to the operators traditional use.

3.4

Overloading binary operators

When a member operator function overloads a binary operator, the function will have only one parameter. This parameter will receive the object that is on the right hand side of the operator. The object on the left hand side is the object that generated the call to the operator function. Here is an example of overloading the assignment operator=().
student &student::operator=(const student &s) { name = s.name; firstname = s.firstname; year = s.year; return *this; }

Note the use of the this pointer and the fact that the operator=() function returns a reference to the object that something is assigned to. This is done in order to allow multiple assignments like a = b = c;, which are syntactically correct in C++. Observe that for basic classes like student the assignment operator is already well dened, similarly to the copy constructor (cf. 2.4.1). This is due to the fact that the built-in assignment operator performs a bit-wise copy operation. Hence in practice one only needs to overload the assignment operator if the class has a pointer to memory that is dynamically allocated. The following program demonstrates that it can be important to dierentiate between the object that generates the call, and the object that is passed as a parameter. In this example the operators -= and are overloaded.
#include <iostream> using namespace std; class Point { double x,y;

15

OOP in C++

Dr Robert Nrnberg u

public: Point(double a = 0, double b = 0) { x = a; y = b; } Point &operator-= (const Point &p) { x -= p.x; y -= p.y; return *this; } Point operator- (const Point &p) const { Point r(x-p.x,y-p.y); return r; } void print() const { cout << x << "," << y << endl; } }; int main() { Point p(1.0,2.0), q(0.0,0.5), r; r = p -= q; p.print(); // 1,1.5 r = p - q; r.print(); return 0; } // 1,1

Observe that the operator-= () function could have been dened to return void instead. But then statements like r = p -= q; would no longer work. 3.4.1 Overloading relational operators

A special family of binary operators are the relational and logical operators. Naturally, they too can be overloaded for newly dened classes. Here is an example.
class Point { double x,y; public: bool operator< (const Point &p) const { return (x < p.x || (x == p.x && y < p.y)); } // etc };

The C++ compiler treats the relational operators like any other binary operator. In particular, overloading the operator < does not give meaning to the operators >, <= and >=. That means that if you want to make use of them, you have to overload them individually.

3.5

Overloading unary operators

Overloading a unary operator is similar to overloading a binary operator. The only dierence is that there is only one operand to deal with. Hence when overloading a unary operator using a member function, the function takes no parameters. Some unary operators have both a prex and a postx version. In order to dierentiate between these two when overloading the operator, the postx version of the operator is dened as if it takes a (dummy) parameter of type int. In the following example you can see how to overload unary operators, and in particular how to dene both the prex and postx version of the ++ operator.
#include <iostream> using namespace std; class Point { double x,y; public: Point(double a = 0, double b = 0) { x = a; y = b; } Point &operator++ () { x++; y++; return *this; } // Point operator++ (int) { Point r(x,y); x++; y++; return r; } // Point operator- () const { Point r(-x,-y); return r; } operator double() const { return x; } // void print(ostream &os) const { os << "(" << x << "," << y << ")"; }; ostream &operator<< (ostream &os, const Point &p) {

prefix postfix casting }

16

OOP in C++

Dr Robert Nrnberg u

p.print(os); return os; } int main() { Point p(0,0); cout << p++ << endl; cout << p << endl; cout << ++p << endl; cout << p << endl; cout << -p << endl; cout << p << endl;

// (0,0) // (1,1) // (2,2) // (2,2) // (-2,-2)

// postfix

// prefix

// always prefix

cout << (double) p << endl; // 2 return 0; }

Note also the example for overloading the type casting operator, in the example above for the casting to type double.

3.6

Overloading the () operator

The () function call operator allows you to dene functions as freely as you can outside classes. The function call operator can take any number of parameters and return any type you want. Here are two simple examples.
#include <iostream> using namespace std; class Point { double x,y; public: Point(double a = 0, double void operator() (double x) cout << "Function called } int operator() () const { cout << "Function called return 1; } }; int main() { Point p; p(2.0); int i = p(); return 0; }

b = 0) { x = a; y = b; } const { with double x=" << x << "." << endl;

with ()." << endl;

// Function called with double x=2. // Function called with ().

3.7

Using friend operators

It is also possible to overload an operator relative to a class by using a (global) friend rather than a member function. In this case, a binary operator has to be passed both operands explicitly, while a unary operator has to be passed the single operand. Friend operator functions are usually used when dening operations involving built-in types, where the built-in type is on the left hand side of the operator. Hence the call to the operator function cannot be generated by an object of your newly dened class. The most common example is to overload the inserter operator <<, where usually an object of type ostream is on the left hand side of the operator. See e.g. the following example.
#include <iostream> using namespace std;

17

OOP in C++

Dr Robert Nrnberg u

class Point { friend ostream &operator<< (ostream &os, const Point &p); friend Point operator- (const double &x, const Point &p); double x,y; public: Point(double a = 0, double b = 0) { x = a; y = b; } }; ostream &operator<< (ostream &os, const Point &p) { os << "(" << p.x << "," << p.y << ")"; return os; } Point operator- (const double &x, const Point &p) { Point r(x-p.x, -p.y); return r; } int main() { Point z(0.5,1.5), r; r = 2.0 - z; cout << r << endl; return 0; }

// access to x and y

// access to x and y

// (1.5,-1.5)

Note, however, that it is not really necessary to use friend operators. See the example in 4.5 for a way to keep the private member data of a class really private.

Inheritance

Inheritance is one of the most important features of C++. It allows classes to inherit properties from another class and hence enables you to build a hierarchy of classes, moving from the most general one to the most specic one. When one class inherits another, it uses this general form:
class derived_class_name : access base_class_name { // etc }

Here, access is one of three keywords: public, private, or protected. The access specier determines how elements of the base class are inherited by the derived class. When the access specier is public, all public members of the base become public members of the derived class. If the access specier is private, however, all public members of the base become private members of the derived class. In either case, all private members of the base remain private to it and are inaccessible by the derived class. Here is a simple example.
class base { public: int a; void print() const { cout << a << endl; } }; class sub : public base { int b; public: void print() const { cout << a << "," << b << endl; } void printbase() const { base::print(); } };

In this example we added the data b to the base denition of the class and redened the print() member function. Observe the usage of the scope operator :: to relate to a base class function with the same name as a (new) function in the derived class. Inheritance from more than one parent class is possible, but is not covered in this course.

18

OOP in C++

Dr Robert Nrnberg u

4.1

Protected members

To allow for private members in the base class that will be private and accessible in the derived class, the access specier protected is provided. It is equivalent to the private specier with the sole exception that protected members of a base class are accessible to members of any class derived from that base. Hence the full general form of a class declaration looks something like this:
class class_name { // private members protected: // optional // protected members public: // public members }

How members of the base class can be accessed in the derived class is shown in the following table. access level : public : protected : private base class members public protected private public protected inaccessible protected protected inaccessible private private inaccessible

Table 1: Accessibility of base class members in derived class.

4.2

Constructors, destructors and inheritance

When a base and derived class both have constructor and destructor functions, the constructor functions are executed in order of derivation. The destructor functions are executed in reverse order. I.e. the base class constructor is executed before the constructor in the derived class. The reverse is true for destructor functions. Furthermore, it is possible for the derived class constructor to pass arguments to the base class constructor. This is done in the following fashion.
class base { protected: int a; public: base(int i) { a = i; } void print() const { cout << a << endl; } }; class sub : public base { int b; public: sub(int i, int j) : base(i) { b = j; } // calls base constructor void print() const { cout << a << "," << b << endl; } };

Note that data members that are inherited from the base class will always be initialized through the base class constructor. This guarantees that even data that is inaccessible in the derived class (cf. 4.1) is properly initialized when an object of the derived class is created.

4.3

Pointers to derived classes

In C++, a pointer declared as a pointer to a base class can also be used to point to any class derived from that base. For example, with the above denitions of base and sub, the following statements are correct:
base *p; // base class pointer base b(1); sub s(1,2); p = &b; p = &s; // p points to base object // p points to derived object

19

OOP in C++

Dr Robert Nrnberg u

Although you can use a base pointer to point to a derived object, you can access only those members of the derived object, that were inherited from the base class. This is because the pointer has knowledge only of the base class. It knows nothing about the members changed or added by the derived class. Note that a pointer to a derived class cannot be used to access an object of the base class. Finally, everything mentioned here about pointers to derived classes also holds true for passing arguments by reference. I.e. a function that expects a reference to a base class object can also be passed a reference to a derived class object. The following example illustrates that.
void show(const base &b); // reference to base class int main() { sub s(1,2); show(s); // etc }

// pass reference to derived object

4.4

Virtual functions

Virtual functions implement the one interface, multiple methods philosophy that underlies polymorphism. A virtual function is a class member function that is declared within a base class and redened by a derived class. The virtual function within the base class denes the form of the interface to that function. Each redenition of the virtual function by the derived class implements its operation as it relates specically to the derived class and hence creates a specic method. In order to declare a function virtual, precede its declaration in the base class with the keyword virtual. What makes a virtual function special and interesting is that it supports so called run-time polymorphism. That is, when a base class pointer (or a base class reference) points to a derived object that contains a virtual function and that virtual function is called by using that pointer, C++ determines which version of that function to call based upon the type of the object that is pointed to. This idea is best illustrated with an example.
#include <iostream> using namespace std; class base { public: virtual void print() const { cout << "base" << endl; } }; class sub : public base { public: void print() const { cout << "sub" << endl; } }; class subsub : public sub { public: void print() const { cout << "subsub" << endl; } }; int main() { base b; sub s; subsub ss; b.print(); s.print(); ss.print(); base *p[3]; p[0] = &ss; p[1] = &s; p[2] = &b; for (int i = 0; i < 3; i++) p[i]->print(); return 0; }

// virtual !

// "base", "sub", "subsub" // pointers to base class // "subsub", "sub", "base"

Observe that without declaring the print() member function in the base class as virtual, the last three outputs would all read "base". Only by making the function virtual do we force the compiler to select the appropriate member function for each object of the derived classes. In case that a virtual function is not overridden in one of the derived classes, the version of the next highest class where the function is redened is executed. E.g. on introducing the derived class 20

OOP in C++

Dr Robert Nrnberg u

class subsubsub : public subsub {};

the following statements in main() would also lead to the output "subsub".
subsubsub sss; p[0] = &sss; p[0]->print();

// "subsub"

Finally, the virtual nature of a base member function is also invoked, when the function is called inside another base member function. Here one can think of the this pointer being used, see 2.6. The this pointer inside a base member function is a pointer of type base *. When that pointer is used to call a virtual member function, C++ determines which version of that function to call, depending on the object it points to. The following example demonstrates this.
#include <iostream> using namespace std; class base { public: virtual void print() const { cout << "base" << endl; } void show() const; }; class sub : public base { public: void print() const { cout << "sub" << endl; } }; void base::show() const { print(); } int main() { base b; sub s; s.show(); b.show(); return 0; } // or: this->print();

// virtual ! // non-virtual

// "sub", "base"

4.4.1

Pure virtual functions

Sometimes when a virtual function is declared in the base class, there is no meaningful operation for it to perform. I.e. its denition in the base class would merely be used to dene the functions interface, while there would be no sensible method for it in the base class. This situation is quite common because often a base class does not dene a complete class by itself. Instead, it simply supplies a core set of member functions and variables to which the derived classes supply the remainder. When there is no meaningful task for the base class virtual function to perform, then any derived class must override this function. In C++ this is ensured by dening the concerned function as pure virtual. A pure virtual function has no denition in the base class, only the functions declaration is included. To make a virtual function a pure virtual function, it has to be assigned the value 0, e.g.
class base { public: virtual void print() const = 0; // etc }

Some programmers do not like this somewhat arbitrary syntax and dene a special preprocessor macro called pure as follows.
#define pure = 0

Then the previous denition of the pure virtual function print() would change to the following.
class base { public: virtual void print() pure; // etc }

21

OOP in C++

Dr Robert Nrnberg u

Keep in mind that when a function is dened as pure virtual, it forces any derived class to provide an overriding denition. 4.4.2 Abstract classes

A class that contains at least one pure virtual function is called an abstract class. All other classes are called concrete classes. Since an abstract class contains at least one function for which no body exists, it is technically speaking an incomplete type. In particular, no objects of that class can be created. The only purpose of abstract classes is to be inherited by derived classes. Note, however, that it is still possible to use pointers (and references) to abstract base classes. Here is an example.
#include <iostream> using namespace std; class base { public: virtual void print() const = 0; };

// pure virtual !

class sub1 : public base { int a; public: void print() const { cout << "sub1: " << a << endl; } }; class sub2 : public base { double b; public: void print() const { cout << "sub2: " << b << endl; } };

int main() { sub1 s1; sub2 s2; base *p[2]; // pointers to base class p[0] = &s1; p[1] = &s2; for (int i = 0; i < 2; i++) p[i]->print(); return 0; }

4.4.3

Constructors and destructors

While constructor functions cannot be made virtual, destructor functions can. The restriction for constructors makes sense since constructors must always be called explicitly, by naming the exact class to construct. They can never be called by pointer or by reference. Moreover there is no need to declare a constructor as virtual, since on recalling 4.2, a constructor in a derived class always invokes the constructor of the next higher class. Destructors, on the other hand, can and often should be made virtual. In particular, if a class contains virtual member functions, then the destructor should be declared virtual to ensure that an object of a derived class is always destructed properly, even when pointed to by a base class pointer. Here is a simple example.
#include <iostream> using namespace std; class base { public: base() { cout << "Constructing base." << endl; } virtual ~base() { cout << "Destructing base." << endl; } virtual void print() const { cout << "base" << endl; } };

// virtual !

22

OOP in C++

Dr Robert Nrnberg u

class sub : public base { public: sub() { cout << "Constructing sub." << endl; } ~sub() { cout << "Destructing sub." << endl; } void print() const { cout << "sub" << endl; } }; int main() { base *p = new sub; delete p; return 0; } // calls ~sub ! // "Destructing sub.", "Destructing base."

4.4.4

Pure virtual destructors

In order to avoid linkage errors at compile time, special care has to be taken when declaring a base class destructor as pure virtual. As the compiler will invoke all the destructors of a given class hierarchy when an object of a derived class runs out of scope, it will in particular also attempt to call the base class destructor. Hence in this particular case, we need to also provide an implementation of the pure virtual destructor member function. As the C++ syntax prohibits this denition inside the class denition, this needs to be done after the class denition. Below is an example.
#include <iostream> using namespace std; class base { public: base() {} virtual ~base() = 0; };

// pure virtual !

/*********************************************************/ /* without the next line, the code does not compile/link */ /*********************************************************/ base::~base() {} class sub : public base { public: ~sub() { cout << "Destructing sub." << endl; } }; int main() { base *p = new sub; delete p; return 0; } // calls ~sub ! // "Destructing sub."

4.4.5

Inheriting virtual functions

When a virtual function is inherited, so is its virtual nature. That means that when a derived class inherits a virtual function from a base class and is then used to derive yet another class, the virtual function may be overridden by the nal derived class. See also the example in 4.4. Moreover, since the inherited function is virtual in the rstly derived class, a pointer to that class can also be used to implement run-time polymorphism. Here is an example.
#include <iostream> using namespace std; class base { public:

23

OOP in C++

Dr Robert Nrnberg u

void print() const { cout << "base" << endl; } // here not virtual! virtual void show() const { cout << "show base" << endl; } }; class sub : public base { public: virtual void print() const { cout << "sub" << endl; } void show() const { cout << "show sub" << endl; } }; class subsub : public sub { public: void print() const { cout << "subsub" << endl; } void show() const { cout << "show subsub" << endl; } };

// now virtual // virtual by default

int main() { base b; sub s; subsub ss; base *pb[3]; sub *ps[2]; pb[0] = &b; pb[1] = &s; pb[2] = &ss; for (int i = 0; i < 3; i++) { // base, show base pb[i]->print(); pb[i]->show(); // base, show sub } // base, show subsub ps[0] = &s; ps[1] = &ss; for (int i = 0; i < 2; i++) { ps[i]->print(); ps[i]->show(); // sub, show sub } // subsub, show subsub return 0; }

4.5

Friend functions and inheritance

A friend function is not inherited. That is, when a base class includes a friend function, that friend function is not a friend of a derived class. A nice way around this problem is to put the main part of the function into a virtual (!) member function. This means in particular, that the function does not need to be declared as friend anymore. The following example demonstrates this using operator overloading, see 3 for details on that.
#include <iostream> using namespace std; class base { protected: int a; public: virtual void print(ostream &os) const { os << a; } }; class sub : public base { int b; public: void print(ostream &os) const { os << a << "," << b; } }; ostream &operator<< (ostream &os, const base& b) { b.print(os); return os; } int main() { base b; sub s; cout << "b = " << b << ", s = " << s << endl; return 0; }

24

OOP in C++

Dr Robert Nrnberg u

4.6

Example for inheritance

The iostream library is an object-oriented library that provides input and output functionality using streams. The following gure illustrates the dependence of the dierent classes provided by the library.

<ios> ios base ? ios  1   

<istream> istream ? iostream 6 Q Q s Q ostream <ostream>

<iostream> - cin cout cerr - clog -

<fstream> ifstream fstream

Q Q

ofstream

Templates

Templates are a very powerful feature of C++. They are a relatively new addition to C++, and they introduce the concept of generic programming. Generic programming is a data structure independent way of developing and delivering algorithms that need not be bound to a specic object. In particular, C++ templates allow you to implement generic functions or classes independent of a certain parameters type. Once a template is dened, it can be used in connection with objects from any type, provided that all the operations within the template are well dened for this type. Templates are very useful when implementing generic constructs like vectors, stacks or lists, which can be used with an arbitrary type. The most famous example for this kind of application is the Standard Template Library, cf. 6. C++ templates provide a way to reuse source code as opposed to inheritance and composition which provide a way to reuse object code. In C++ you can dene two kinds of templates: class templates and function templates. The Standard Template Library, for example, uses function templates to dene generic algorithms, and the containers are implemented as class templates. More details on templates and many examples can be found at www.josuttis.com/tmplbook.

5.1

Function templates

Here is a very simple example that demonstrates the use of function templates. Observe that the syntax template <class T> is equivalent to the syntax template <typename T> used in this example.
#include <iostream> using namespace std; template <typename T> T absolute(const T &a) { if (a < 0) return -a; else return a; } int main() { int a = -5; double b = 1.2; float c = -0.43; long int d = -123456789; cout << "Absolute values are : " << absolute(a) << "," << absolute(b) << "," << absolute(c) << "," << absolute(d) << endl; // the following function uses the type T

25

OOP in C++

Dr Robert Nrnberg u

return 0; }

Note that for the above program the compiler creates four dierent versions of the function absolute() one for each of the four types the function is invoked for. Moreover, it is worth noting that one could have referred to the dierent overloaded versions directly, as in the following example.
cout << "Absolute values are : " << absolute<int>(a) << endl;

In the above example, this is not really necessary. However, this way of identifying the intended function has to be used, whenever at least one template typename T does not appear in the signature of the function. Here is a small example.
#include <iostream> using namespace std; template <typename T> void foo(int i) { T local = 0; cout << local << " " << i << endl; } int main() { foo<double>(2); return 0; }

5.2

Class templates

Similar to function templates, a class template denition looks like a regular class denition, except that it is prexed by the keyword template. For example, the next program denes a class template my_pair.
#include <iostream> using namespace std; template <typename T> class my_pair { T a,b; public: my_pair(T i, T j) : a(i), b(j) {} void print() const; }; template <typename T> void my_pair<T>::print() const { cout << a << "," << b << endl; } int main() { my_pair<int> a(-1,2); my_pair<double> b(1.2,0.1); a.print(); b.print(); return 0; }

// define print for my_pair<T>

5.2.1

Inheritance

When inheriting from a class template, one has to specify which version of the base class one refers to. In particular, one could either inherit the full template, or just one instantiation. Here is an example for both.
#include <iostream> using namespace std; template <typename T> class base { // base template class

26

OOP in C++

Dr Robert Nrnberg u

protected: T a; public: base(T i = 0) : a(i) {} void print() const { cout << a << endl; } }; template <typename T> class sub : public base<T> { // refer to base<T> , protected: T b; public: sub(T i, T j) : base<T>(i), b(j) {} // its constructor , void print() const { cout << base<T>::a << " " << b << endl; } // and its data }; class subsub : public sub<int> { // refer to sub<int> int c; public: subsub(int i, int j, int k = 0) : sub<int>(i,j), c(k) {} void print() const { cout << a << " " << b << " " << c << endl; } }; int main() { subsub sub<double>

a(-1,2); b(1.2,0.1);

a.print(); b.print(); return 0; }

Here a short remark is due. Observe that in the member functions of the inherited class template sub<T> particular attention has to be paid on how to refer to member data that was inherited from base<T>. In this example, the correct syntax is
base<T>::a

which is slightly cumbersome. However, this is the only way we can refer to that member data. The reason behind that lies in the C++ standard for templates. Strictly speaking, when the member function in sub<T> is compiled, the base class itself has not yet been instantiated and as a consequence we have no knowledge of what member data it will provide. Referring to the desired member data directly with the help of the scope operator :: circumvents this problem. 5.2.2 Template specialization

A template specialization allows a template to have specic implementations when the parameter type is of a determined type. For instance, in the following example the member function mod() returns the module between two integers if the class is my_pair<int> and 0 otherwise.
#include <iostream> using namespace std; template <typename T> class my_pair { T a,b; public: my_pair(T i, T j) : a(i), b(j) {} T mod() { return 0; } }; template <> int my_pair<int>::mod() { return a % b; } int main() {

// overload definition for my_pair<int>

27

OOP in C++

Dr Robert Nrnberg u

my_pair<int> my_pair<double> cout << a.mod() cout << b.mod() return 0; }

a(-1,2); b(1.2,0.1); << endl; << endl;

// -1 // 0

by default

Of course, we can also use template specialization in order to identify the typename that our template was instantiated for. The following example demonstrates this.
#include <iostream> using namespace std; template <typename T> class my_pair { T a,b; public: my_pair(T i, T j) : a(i), b(j) {} int type_id() const { return 0; } void print() const; }; template <> int my_pair<int>::type_id() const { return 1; } template <> int my_pair<double>::type_id() const { return 2; } template <typename T> void my_pair<T>::print() const { int t = type_id(); switch (t) { case 1 : cout << "Two integers: "; break; case 2 : cout << "Two doubles: "; break; default: cout << "Unknown type: "; break; } cout << a << " , " << b << endl; } int main() { my_pair<int> a(-1,2); my_pair<double> b(1.2,0.1); my_pair<float> c(-1.1,0.2); a.print(); b.print(); c.print(); return 0; } // Two integers: -1 , 2 // Two doubles: 1.2 , 0.1 // Unknown type: -1.1 , 0.2

// used to identify T

// specialize for int

// specialize for double

5.2.3

Partial specialization

It is also possible to generate a specialization of the class for just one of several type parameters. Here is an example.
#include <iostream> using namespace std; template <typename T1, typename T2> class my_pair { T1 a; T2 b; public: my_pair(T1 i, T2 j) : a(i), b(j) {} void print() const; }; // base template class

28

OOP in C++

Dr Robert Nrnberg u

template <typename T1, typename T2> void my_pair<T1, T2>::print() const { cout << a << "," << b << endl; } template <typename T> class my_pair<T, double> { T a; double b; public: my_pair(T i, double j) : a(i), b(j) {} void print() const; }; // partial specialization

template <typename T> void my_pair<T, double>::print() const { cout << "The second data member is of type double: " << a << "," << b << endl; } int main() { my_pair<int, int> a(-1,2); my_pair<double, double> b(1.2,0.1); a.print(); b.print(); return 0; }

Note that the denition of any specialization is independent from the base template. I.e. in theory a specialization could implement very dierent methods, though this would be very bad programming practice indeed.

5.3

Compile time computations

Templates in C++ do not only oer compile time polymorphism as described above. One of the less exploited and rather theoretical features is compile time computation. The following example computes factorials of constant integer expressions at compile time.
#include <iostream> using namespace std; template <int N> class Factorial { public: enum { value = N * Factorial<N-1>::value }; }; template <> class Factorial<0> { public: enum { value = 1 }; }; int main () { cout << "10! = " << Factorial<10>::value << endl; return 0; }

The beauty of the above example is that no computation time is needed at run time.

STL

The Standard Template Library (STL) is a general purpose C++ library of algorithms and data structures. The STL is part of the standard ANSI-C++ library and is implemented by means of the C++ template mechanism, hence its name. While some aspects of the library are very complex, it can often be applied

29

OOP in C++

Dr Robert Nrnberg u

in a very straightforward way, facilitating reuse of the sophisticated data structures and algorithms it contains. For a full documentation see www.sgi.com/tech/stl.

6.1

Vectors

The vector template is probably the most important and most widely used container class oered by the STL. What makes it so attractive is that one can use it in the same fashion as C style arrays without having to worry about memory management or possible memory leaks. Furthermore, by virtue of the size() member function, one knows at any point in time how many objects a vector contains. With C style arrays, of course, failure to monitor the size of an array properly can lead to incorrect programs and segmentation faults/fatal exception errors. All of the STL container classes are treated by C++ almost like the built-in standard types. In particular, C++ takes care of the proper destruction of vectors and other containers as soon as one of these objects goes out of scope. Moreover, for the vector class basic operators like =, == and < are already dened and functional. The following table gives an overview over the vector classs most important member functions. Here we assume that the size of the vector is of type int, that the objects contained in the vector are of type T and that the iterators are of type T*. Some of the type declarations will be omitted in the table. Details can be found at www.sgi.com/tech/stl/Vector.html.
Member function begin() end() size() capacity() bool empty() operator[] (int i) at(int i) vector() vector(int n) vector(int n, const T &t) vector(const vector&) reserve(int n) push_back(const T&) pop_back() swap(vector&) insert(T *p, int n, const T& t) erase(T *p) erase(T *f, T *l) clear() resize(int n, const T &t = T()) operator= (const vector&) operator== (const vector&) operator< (const vector&) Description Returns iterator pointing to the beginning of vector. Returns iterator pointing to the end of vector. Size of vector. Number of elements for which memory has been allocated. True if vectors size is 0. Returns reference to or the nth element. Same as [] but with boundary check. Not always available. Default constructor: creates an empty vector. Creates a vector with n elements. Creates a vector with n copies of t. Copy constructor. Increase capacity() to n. Inserts a new element at the end. Removes the last element. Swaps the contents of two vectors. Inserts n copies of t before p. Erases the element at position p. Erases range [f, l). Erases all of the elements. Inserts/erases elements at the end so that size becomes n. Assignment operator. Tests two vectors for equality. Lexicographical comparison. Example T *p = v.begin(); T *p = v.end(); if ( i < v.size() ) if (!v.empty() ) v[2] = v[3]; v.at(2) = v.at(3); vector<T> v; vector<T> v(100); vector<T> v(100, t); vector<T> u(v); v.reserve(10000); v.push_back(5); v.pop_back(); u.swap(v); u.insert(u.end(), 9, t); v.erase(v.begin() ); v.erase(&v[3], &v[8]); v.clear(); v.resize(50); u = v; if (u == v) if (u < v)

Table 2: Member functions of the STL vector container class. Strictly speaking, the last two functions are no member functions. Instead they are implemented as global functions, taking two parameters of type const vector&.

6.2

Other sequential containers

The vector container is one of the sequence containers that the STL oers. The other sequence containers are deque and list. Note that the vector and deque classes support random access to elements, while the list class does not. That is why the STL sort() algorithm (cf. 6.6) does not work for the list class. Hence a sorting algorithm for lists is provided as a member function, see the example below. The following table shows the insert and erase overheads of the containers vector, list and deque. You should keep these in mind when choosing a container for solving a specic task. E.g. a list is preferable when you do not need random access to elements and when you often have to erase elements in the middle, say. Here is a simple example for the list container class. Note that iterators will be covered in more detail in 6.4.
#include <iostream> #include <list> using namespace std;

30

OOP in C++

Dr Robert Nrnberg u

container vector list deque

insert/erase overhead at the beginning in the middle at the end linear linear constant constant constant constant constant linear constant

Table 3: Insert and erase overheads for vector, list and deque.

int main() { list<int> L; L.push_back(0); L.push_front(0); L.insert(++L.begin(),2,3); L.push_back(5); L.push_back(6); L.sort(); // the STL sort() algorithm does NOT work // linear traversal

for(list<int>::iterator i = L.begin(); i != L.end(); ++i) cout << *i << " "; cout << endl; return 0; }

A second group of STL containers are the container adaptors: stack, queue and priority_queue. They were derived from the three original containers described above.

6.3

Associative containers

The last group are the so called associative containers. They include set, multiset, map and multimap. A map, for example, is a sorted associative container that holds pairs of keys and data, where each entrys key is unique. Hence it is ideally suited to implement dictionaries, for instance. Here is a short example for the map container class.
#include <iostream> #include <map> #include <string> using namespace std; int main() { map<string, int> freq; string word; while (cin >> word && word != "quit") { ++freq[word]; } // write the frequency and the word for (map<string, int>::iterator i = freq.begin(); i != freq.end(); ++i) { cout << i->second << " " << i->first << endl; } return 0; }

// map of words and their frequencies

Details on all the STL containers can be found at the web site given before and also at e.g. www.cppreference.com.

31

OOP in C++

Dr Robert Nrnberg u

6.4

Iterators

Iterators are a fundamental feature of the Standard Template Library. The concept of iterators was introduced in order to unify access to and traversal of dierent kinds of STL containers and, in particular, to allow most of the STL algorithms to work for any container type. Let us look at a simple example.
#include <iostream> #include <vector> #include <list> #include <map> #include <string> using namespace std; int main() { vector<double> v(20,1.2); list<int> l(10,4); map<string, int> m; m["ab"] = 1; m["yz"] = 4; m["fg"] = 0; map<string, int>::iterator mitr; for(mitr = m.begin(); mitr != m.end(); ++mitr) cout << (*mitr).first << " " << (*mitr).second << endl; list<int>::iterator litr; for(litr = l.begin(); litr != l.end(); ++litr) cout << *litr << " "; cout << endl; vector<double>::iterator vitr; for(vitr = v.begin(); vitr != v.end(); ++vitr) cout << *vitr << " "; cout << endl; return 0; }

Observe that the traversal of e.g. a map, a list and a vector can be done in exactly the same way, although the internal data structures of these containers are fundamentally dierent. You can think of iterators as a special class that belongs to the STL containers and that is overloaded dierently for each container. Through appropriately overloading operators such as =, !=, ++, * (see above) and other operators, this iterator class provides access to the elements stored in a container and allows its traversal. Note that pointers are nothing other than special iterators for both C style arrays and STL vectors. There are several advantages for the use of iterators. As you have seen, iterators can be uniformly used for all containers and also arrays. Furthermore, on creating your own container class with an appropriate iterator class, you can use the existing STL algorithms for your new class. Similarly, if you code a new (template based) algorithm for container classes that is written using iterators, it will apply to all existing containers for which there are iterators. More details on iterators can be found at www.sgi.com/tech/stl/Iterators.html.

6.5

Functors

Many of the STL algorithms (cf. 6.6) can be given so called function objects, or functors, as one of their parameters. A functor is simply any object that can be called as if it is a function. Hence an ordinary function is a functor, and so is a function pointer. Moreover, any object of a class that denes an operator() is a functor. The STL oers many predened generic functors, such as plus, minus, multiplies, divides and modulus. Furthermore, many bool functors so called predicates are dened within the STL as well. E.g. equal_to, not_equal_to, greater, less, greater_equal and less_equal. The following program shows how some of these templates can be used.
#include <iostream> #include <cmath> #include <functional>

32

OOP in C++

Dr Robert Nrnberg u

using namespace std; struct abs_greater : public binary_function<double, double, bool> { bool operator()(double a, double b) { return (abs(a) > abs(b)); } }; int main() { double a = -5.1, b = 3.4; abs_greater a_g; plus<double> pl; if (a_g(a,b)) cout << "|a| > |b|" << endl; cout << "a + b = " << pl(a,b) << endl; return 0; }

Of course, as stand alone classes these functors do not make much sense. But in connection with the STL algorithms, they can be very powerful.

6.6

Algorithms

The STL also includes a large collection of algorithms that manipulate the data stored in containers. The following example demonstrates the use of some of the STL algorithms. Although the program uses a vector<int> container, most of the algorithms also work for many other STL containers. To make this possible, all the algorithms are dened as global functions. Secondly, most of the algorithms are given a range to operate on. E.g. a full vector<int> v is given by the range (v.begin(), v.end()).
#include <iostream> #include <vector> #include <algorithm> #include <numeric> #include <iterator> using namespace std; int dice() { return rand() % 6 + 1; } int main() { vector<int> v(1000); generate(v.begin(), v.end(), dice); // generate cout << "Number of sixes: " << count(v.begin(), v.end(), 6) << endl; // count vector<int>::iterator firstfive = find(v.begin(), v.end(), 5); cout << "Number after first five = " << *(firstfive+1) << endl; reverse(v.begin()+10, v.begin()+100); random_shuffle(v.begin(), v.end()); sort(v.begin(), v.end()); sort(v.begin(), v.end(), greater<int>() ); double a = max(1.2, 4.3); cout << "max(1.2,4.3) = " << a << endl; vector<int>::iterator min_el = min_element(v.begin(), v.end()); cout << "The smallest element is " << *min_el << endl; vector<int> u(10); for (unsigned int i = 0; i < u.size(); ++i) u[i] = i - 4; while (binary_search(u.begin(), u.end(), 0)) { cout << "There exists a 0 element." << endl; vector<int>::iterator zero = find(u.begin(), u.end(), 0); u.erase(zero); // binary_search // find

// reverse // random_shuffle // sort with < // sort with greater // max;

// min_element

33

OOP in C++

Dr Robert Nrnberg u

} vector<int>::iterator new_end = remove(u.begin(), u.end(), -2); u.erase(new_end, u.end()); copy(u.begin(), u.end(), ostream_iterator<int>(cout, " ")); cout << endl; // remove

// copy // -4 -3 -1 1 2 3 4 5

cout << "The sum of all elements is " << accumulate(u.begin(), u.end(), 0) << endl; // accumulate cout << "The product of all elements is " << accumulate(u.begin(), u.end(), 1, multiplies<int>()) << endl; return 0; }

The remarkable thing about the STL algorithms is that they work with almost any kind of container, including C style arrays. Here is an example, that rst sorts and then outputs a given array.
#include <iostream> #include <algorithm> #include <iterator> using namespace std; int main() { int A[] = {1, 4, 2, 8, 5, 7}; const int N = sizeof(A) / sizeof(int); sort(A, A + N); copy(A, A + N, ostream_iterator<int>(cout, " ")); cout << endl; return 0; } // sort // copy : "1 2 4 5 7 8"

7
7.1

Miscellaneous
Namespaces

Namespaces allow you to group a set of global classes, objects and/or functions under a name. This can be useful when writing large programs, where one wants to use the same name for two dierent objects in dierent contexts, say. If one chooses not to use the using namespace keyword, one can still refer to the dened objects with the help of the scope operator ::. Here is an example.
#include <iostream> using namespace std; namespace first { int var = 5; } namespace second { double var = 3.1416; } int main () { { using namespace first; cout << var << endl; } { using namespace second; cout << var << endl; } cout << first::var + second::var << endl; return 0; }

When using the ANSI-C++ compliant include les one has to bear in mind that all the included functions, classes and objects will be declared under the std namespace. 34

OOP in C++

Dr Robert Nrnberg u

7.2

Advanced I/O
#include <iostream> using namespace std; int main() { cout << 123.23 << endl; // 123.23 cout.setf(ios::scientific | ios::showpos | ios::uppercase); cout << 123.23 << endl; // +1.232300E+02 cout.unsetf(ios::showpos | ios::uppercase); cout << 123.23 << endl; // 1.232300e+02 cout.precision(1); cout << 123.23 << endl; // 1.2e+02 cout.precision(5); cout.width(13); cout << 123.23 << endl; // 1.23230e+02 cout.precision(5); cout.width(13); cout.fill(0); cout << 123.23 << endl; // 001.23230e+02 cout.precision(3); cout.width(13); cout << 123.23 << endl; // 00001.232e+02 cout << 123.23 << endl; // 1.232e+02 return 0; }

Here is an example that demonstrates some of the advanced features of output operations under C++.

Of interested could also be the ags ios::right and ios::fixed. For more details visit www.cppreference.com/cppio.html. Another convenient way to simplify I/O operations is to dene your own manipulators. Here is a short example.
#include <iostream> using namespace std; ostream &myformat(ostream &os) { os.width(10); os.precision(4); os.fill(*); return os; } ostream &tri(ostream &os) { os << "The result is: "; return os; } int main() { cout << tri << myformat << 123.23 << endl; return 0; }

// The result is: *****123.2

7.3

Static class members

If a class member is declared as static, only one copy of it is ever created, no matter how many objects of that class are created. In particular, the programmer has to dene the variable also outside the class, so that memory is allocated for it. The following example demonstrates this.
#include <iostream> using namespace std; class myclass { public: static int i; myclass(int ii) { i = ii;} }; int myclass::i = 10;

35

OOP in C++

Dr Robert Nrnberg u

int main() { myclass a(5); myclass b(8); cout << a.i << endl; return 0; }

// 8

7.4

Conversion functions

In C++ it is possible to dene a conversion function that is used whenever an object of this class is found in a place where a dierent type is expected. Usually this will be one of the built-in C++ types. Here is an example.
#include <iostream> #include <cmath> using namespace std; class cmplx { double r, i; public: cmplx(double a, double b) : r(a), i(b) {} operator double() { return sqrt(r*r + i*i); } }; int main() { cmplx c(2.0,1.5); double a = c; cout << a << endl; return 0; }

// conversion to double // 2.5

Obviously, one should use these conversion functions carefully.

7.5

Exception handling

Exception handling can be a useful tool when developing and debugging a program. The three commands used for it are try, throw and catch.
#include <iostream> #include <string> using namespace std; int main () { try { double *m; m = new double[10]; if (m == NULL) throw string("Allocation failure"); for (int i = 0; i < 100; i++) { if (i > 9) throw i; m[i] = i * i; } // etc delete [] m; } catch (int n) { cout << "Exception: index " << n << " is out of range." << endl; return 1; } catch (string &s) { cout << "Exception: " << s << endl; return 2; } return 0; }

36

OOP in C++

Dr Robert Nrnberg u

If you want to catch any exception within a try block, use catch(...) (i.e. three single dots within the brackets). Finally note that exceptions can also be thrown by functions called within a try block. Below is a simple example.
#include <iostream> #include <string> using namespace std; double ratio(double a, double b) { if (b == 0.0) throw string("Division by zero"); return a / b; } int main () { try { while (1) { double x, y; cout << "x, y = "; cin >> x >> y; cout << " x / y = " << ratio(x,y) << endl; } } catch (string &s) { cout << "Exception: " << s << endl; return -1; } return 0; }

7.6

Temporary objects

In C++ temporary objects are not only created when functions return objects and when functions are passed objects by value. C++ also allows you to dene your own temporary objects of your newly dened classes. These temporary objects are only valid for the line of code their appear in. Here is a simple example.
#include <iostream> #include <vector> using namespace std; class A { int a,b; public: A(int i = 0, int j = 0) : a(i), b(j) {} void print() const { cout << a << "," << b << endl; } }; void show(const A &x, const A &y) { cout << "x = "; x.print(); cout << "y = "; y.print(); } int main () { show(A(), A(1,2)); vector<A> v(10, A(4,8)); for (unsigned int i = 0; i < v.size(); i++) v[i].print(); return 0; }

Note that in the above a temporary object is used to initialize a vector<A>.

7.7

Passing functions as arguments

On recalling 3.1, it is straightforward to pass functions as arguments to other functions. Here is an example. 37

OOP in C++

Dr Robert Nrnberg u

#include <iostream> using namespace std; double abs(double x) { return x > 0 ? x : -x; } void print(double x) { cout << "Number is: " << x << endl; } int combine(void (*f)(double), double (*g)(double), double x) { if (f != NULL && g != NULL) { f( g(x) ); return 0; } else return 1; } int main() { return combine(print, abs, -2.5); }

See also 1.7.1 for another example.

38